Testing different convolution engines

[Keith_W]

Perhaps you might like to repeat the experiment. Do you have the tools you need, i.e. some way to route the output of the convolver to REW? You can download a copy of Acourate Convolver and use it for free for 14 days: https://www.audiovero.de/en/downloads.php

[pkane2001]

27 minutes ago, Keith_W said:

Perhaps you might like to repeat the experiment. Do you have the tools you need, i.e. some way to route the output of the convolver to REW? You can download a copy of Acourate Convolver and use it for free for 14 days: https://www.audiovero.de/en/downloads.php

 

I'll have to think of a way to route the output of HQP convolver to REW. Not immediately obvious to me how to do that, but with all the USB interface boxes and AD/DA converters, I would think there should be a way.

EDIT: Ah, should be pretty easy. I have a Macbook that has digital audio (Toslink) input and output. Just need to connect them together.

[Bob Stern]

Miska uses the convolution himself, so I expect he'd be eager to verify and fix any bug.

 

Keith, I suggest you bring this to Miska's attention by posting a link to this thread in the main HQP thread:

https://www.computeraudiophile.com/forums/topic/19715-hq-player/?page=444

 

[Miska]

Where is the result with Acourate Convolver?

 

To say anything, I'd like to know source content rate, convolution engine settings, filter sampling rate, and output rate... Without knowing the exact use case it is hard to say anything.

[Miska]

5 hours ago, Keith_W said:

OK I managed to superimpose the two images using Photoshop. The thick black line is the summation of the crossover files in Acourate. 

 

 

At least here HQPlayer result looks much better (flatter) than the JRiver result.

 

[Bob Stern]

8 hours ago, Keith_W said:

max out the RMS volume at -3.0.

 

Suggestions:

 

1.  Disable the "Add timing reference signal" option.

 

2. Reduce the level to –10 dB to reduce risk of clipping.

 

3.  When you load the filter in HQP, what value is displayed in the "IR Gain" field?  A positive number means the filter has gain, which leads to clipping.

 

4.  Does the volume knob in HQP change color while playing the test signal, indicating a level near clipping?

 

(Edit:  I posted this without realizing Miska already had replied.  I defer to his obviously more expert advice!)

[Miska]

Those are all good suggestions too...

[Keith_W]

2 hours ago, Miska said:

 

 

 

I posted the same question on the Acourate forum. You can't view that thread without registering first, so for everyone's benefit this is a copy and paste of Uli's answer: 

 

(edit) trying to fix the funky quotes but I can't seem to! 

 

Quote

you can check the AC the following way:

 

With your multiway system Acourate also creates the files CorTestL48.dbl and CorTestR48.dbl. The file contains the full range correction filter. Load them and and save them as CPV files (select the according extension in the Save dialog).

Start the AC and prepare a setup with analog out1/2 and analog in 1/2 and the CPV filters. 

Now open the Acourate logsweeprecorder and select AcourateAsio as driver. Select the output channels 1+2 and the input channel 1 (the input 1 is forced by the AcourateAsio driver)

Open Totalmix and select a loopback for channels 1+2.

Run a logsweep recording and check the result of the sweep by comparing it to the CorTest filter.

 

A long time ago I have verified the convolution to be accurate.

 

And another reply, 

 

 

 

 

at 70 and 650Hz seems to be crossover frequency.

 

Have you measured once with a  "null filter" (a dirac). should give a flat curve. Just to check the HW/SW setup

 

I did the same with a 4-way crossover a while ago (RME fireface UFX with AC 1.5.3 I guess, now use AC1.5.4) and the deviation was less than a tenth of a dB.

Used acourateASIO as driver, Win7Pro, JRiver MC22, 44.1kHz, 65000taps.

 

HQ Player I never used, so can not comment.

 

 

Both very good suggestions. 

 

I have not ruled out user error as the reason for these measurements sounding and looking different. I will repeat the experiment with the suggestions made by Uli. Miska and Bob I will get back to you later today. 

 

 

 

[pkane2001]

7 hours ago, pkane2001 said:

 

I'll have to think of a way to route the output of HQP convolver to REW. Not immediately obvious to me how to do that, but with all the USB interface boxes and AD/DA converters, I would think there should be a way.

EDIT: Ah, should be pretty easy. I have a Macbook that has digital audio (Toslink) input and output. Just need to connect them together.

 

@Keith_W:

 

I ran the test with HQP convolver and digital loopback into REW. I don't get the results you did, no matter what I try. The convolved output after HQP matches exactly the impulse response curve, there's no difference at all. Sweep was recorded at 96/24 and then played back and captured at the same rate.

 

Here's what it looks like in my testing (black line is my measured convolved output after HQP, the blue line above is the impulse response file used for convolution, the red line is your HQP output and the green live is JRiver/Acourate:

 

 

 

[Keith_W]

Thank you Paul. I repeated the experiment, this time with no corrections (null filter / Dirac). 

 

Once again: 

Setup 1:

- JRiver output 48kHz, Volume MAX

- Acourate Convolver: ISO 226 OFF, Flow OFF, Clean OFF, EQ OFF, Dither OFF, Volume MAX. 

 

Setup 2: 
- HQPlayer output 48kHz PCM, Volume MAX

 

Input volume was set using RME Totalmix to make sure there was no clipping. Result: 

 

 

This time there is no difference between HQP and AC apart from the volume difference which I deliberately introduced to stagger the two curves. The reason neither sums to a flat line is because I stupidly forgot to remove the microphone calibration from REW prior to taking the sweep. This probably explains why neither of the curves I posted last night matches the predicted curve, but it does not explain the difference between the convolution engines. 

 

I think the result of this particular test tells me that HQP and AC will output exactly the same signal. The difference in the convolution engines I could hear and measure is probably the result of a difference between the filters, rather than the convolution engine. In other words, I must have made a mistake somewhere when it came to generating the filters. 

 

Call it a novice mistake. I will have to redo the whole test.